---
id: "CVE-2026-116172"
kind: "vulnerability"
aliases: ["CVE-2026-116172"]
title: "vLLM: oversized top_n in gRPC requests crashes the inference engine"
severity: "high"
cvss: 8.2
epss: null
exploited: false
status: "open"
category: "ai-tooling"
products: []
ecosystems: []
tags: ["ai-inference", "ai-tooling", "grpc", "self-hosted", "vllm"]
not_affected_if: ["default-config", "internal-network-only"]
published_at: "2026-10-25T14:16:56+00:00"
first_seen_at: "2026-10-25T14:41:23+00:00"
updated_at: "2026-10-25T14:42:26+00:00"
sources: [{"source": "nvd", "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-116172", "first_seen_at": "2026-10-25T14:41:23+00:00"}]
---

# vLLM: oversized top_n in gRPC requests crashes the inference engine

vLLM versions 0.22.0 before 0.24.0 fail to validate the top_n field in the Rust gRPC converter, so the max_logprobs limit can be bypassed. A remote attacker sending Generate or GenerateStream requests with oversized or integer-wrapping top_n values makes torch.topk fail and terminates EngineCore, which takes the model server down.

_Summary, check and fix written by a model from the sources; labelled as such._

## How to check

- Run `pip show vllm` and confirm whether the version is at least `0.22.0` and below `0.24.0`.
- Check whether the gRPC server is exposed: look for the gRPC serving entrypoint and whether its port is reachable from outside the network.

## Fix

- Upgrade vLLM to `0.24.0` or later.
- Until upgraded, block external access to the gRPC port.

## Timeline

- 2026-10-25T14:41:23+00:00 nvd https://nvd.nist.gov/vuln/detail/CVE-2026-116172

## From the sources

vLLM from 0.22.0 before 0.24.0 contains an improper input validation vulnerability in the Rust gRPC converter that allows remote attackers to bypass max_logprobs limits via CandidateTokens.top_n. Attackers can send Generate or GenerateStream requests with oversized or integer-wrapping top_n values to trigger torch.topk failures that terminate EngineCore.

---

_Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai_